The Seminar Series hosted Professor Roland Sturm from the RAND Corporation at the end of April. The economist, who has worked with big data for three decades, spoke to em.muni.cz about his work and the current situation in American science.
You are an expert in statistics and econometrics, why did you choose these fields?
I liked statistics in high school. I wasn't very good at math, but the statistics we got to at the end of my studies intrigued me. Since then, I've enjoyed it, and I also enjoy working with data because it helps us find out things about the real world.
How does working with data help us understand the world?
Statistics is a very important tool for people because there is a huge amount of data around us that we cannot process with just the normal human mind. It is statistics that allows us to extract the information we want and need from big data. It's a great tool that I've been working on for half a century.
What exactly is big data and is it all relevant for research purposes?
The definition of "big data" has changed over the years. What used to be considered big data, such as the operational records of nuclear power plants, would now be a tiny amount of information. In terms of relevance, I like data generated for specific purposes that contain real information. This includes, for example, the aforementioned operational records, supermarket food purchases, or medical records. I don't find social media data very useful, even though it is readily available. For example, more than a decade ago, the idea of tracking the flu was conceived by observing when people search for symptoms of the disease. The idea failed very quickly because it's not real data - information seeking is not the same as a medical report or test result.
How has technological progress affected your ability to work with data?
It has been significant. The computers I had thirty years ago couldn't handle that much data. What we can do today on a laptop used to require a supercomputer. The improvement in data storage and processing speed is tremendous. Of course, for more complex models, larger computers are still required.
What are the main challenges associated with big data suitable for research purposes?
Apart from the fact that a lot of the data is not collected at all, it is access to the data. This is very difficult everywhere in the world. Often, it exists at the country level but it is difficult to access it; it requires a lot of work, networking and persuasion. Although it is often not exact data from a laboratory, it is of better quality than that questionnaires or interviews with a small number of respondents would provide. There are also data primarily collected for the business needs of companies, and these are also hard to obtain. Researchers should therefore try to source and build their own data sources.
What big data research have you worked on in recent years?
For example, one project was analysing billions of purchases in 500 supermarkets in South Africa. It was for an insurance company providing discounts on healthy food. Currently, I'm working on predicting health complications caused by diabetes. We've been tracking doctor visits and test results for 170,000 people for several years.
You are heavily involved in health issues, what made you do it?
My interest in public health comes from a general interest in learning about the world. However, I didn't choose it on purpose, it's just that when I joined the RAND Corporation, I took over the projects of retiring economists, and they were health-related. But it was an opportunity to work on something interesting, and so I'm now considered more of a health economist. But I'm happy to study anything where I can get good data from which I can learn something new.
In addition to working with big data, you are also involved in microsimulations. Can you explain this method?
Microsimulations are a way to make economic predictions and forecasts. They originated in the 1950s. They model the behaviour of individuals - people, families, firms - and the interactions between them. We can find out much more information about the relationships between these units, summarise them and create aggregate data to help model trends in the economy. However, this method is very computationally intensive, so it is only with technological progress that it is becoming a very useful tool. Examples of its use include estimating the impact of ageing and associated diseases on the labour market and forecasting the expected total cost of healthcare.
Does your research have any direct impact on politics, for example?
As scientists, we can disseminate the results of our research, for example, through briefings with decision-makers, or we can organise seminars for legislators. But my work is mostly only disseminated through publications and collaboration with journalists. And I have to say that some of my most cited papers have even been based on inquiries from journalists.
Speaking of politics, what is your take on the current situation in American science in the context of the actions of the new administration and president?
There is a lot of uncertainty and frustration about the future. It is very difficult to assess the long-term impact of specific measures because the situation changes every day. For example, as far as I am concerned, I have a grant where I don't have funding at the moment, so the work is delayed. It will probably not kill the project, but it will certainly delay it. How much, I can't say. It's just that uncertainty is the key word in American science at the moment.
In your opinion, has there been any lasting damage to science and research in the US?
There are fields that are certainly being negatively impacted by the current situation. For example, experts are leaving some organisations and this 'brain drain' will have long-lasting consequences. Of course, science in the US is not destroyed, but it is damaged. But how great or significant the damage is cannot be determined now.